W._Michael_Terenyi@QZCOM.MAILNET (10/14/86)
> Faced with a closely-related problem (TeX DVI files and a PASCAL > DVI-to-Impress program), I avoided the issue. I wrote modules that > replaced the PASCAL I/O calls. These were FORTRAN (yes FORTRAN) > subroutines which mapped the DVI file as a section, and then > handled it as a simple vector of bytes along with a pointer to > the current location. > > I realize it may sound complicated, but the system services needed > to map the file aren't really very tough. Your source (512 byte, > fixed length, unformatted data) is an easy target to map to since > # of records <--> number of pages, and there should be nothing in > the file that isn't part of the byte stream. I think you'll find > it significantly faster than ASCIIfying the DVI and then back- > converting to binary. Although this solution seems to be elegant it has to be pointed out that all I/O of the section file is processed through the secondary caches (modified and free list) which has a severe impact on the overall performance due to the increased paging rates. Michael.